GH4 Leeming LUT One – the best LUT for the Panasonic GH4

Thomas Smet said:
Agreed. I also prefer v201. Showed a few tests to a friend and he also preferred skin tones in v201. He said v308 made me look sun burnt.
Click to expand...
I can't win :D Some people complained about the v201 being too blue (they were right, even though a white balance fixed it). Now people say it's too orange :D

Actually, I can understand why someone would prefer the initial output from v201, as it is a touch more blue in the general tone, making skin tones 'pop' more off the background.

I'm releasing another update in the next day or two, v323, which raises the skin tone luma to be more accurate. The colours are accurate with v308, but skin luma is a bit low compared to where it should be, hence the update.

Hopefully then the 'sunburnt' problem will be no more!

Cheers,

Paul :)
 
I've also just released a LUT for the JVC GY-LS300, thanks to Lee Powell (whom a lot of you will know for his excellent work on the GH2 hacks).

This is a pretty awesome camcorder form-factor, M43 mount, Super35 sensor 4K camera that has really impressed me with its log implementation, using the FULL 8 bit depth of the curve instead of the constrained curves of Panasonic and Sony.

The LUT makes J-Log1 a breeze to use and I hope this will spur some renewed interest in the camera for those who want the abilities of a camcorder with a cinematic, interchangeable lens look.

Cheers,

Paul :)
 
visceralpsyche said:
I can't win :D Some people complained about the v201 being too blue (they were right, even though a white balance fixed it). Now people say it's too orange :D

Actually, I can understand why someone would prefer the initial output from v201, as it is a touch more blue in the general tone, making skin tones 'pop' more off the background.

I'm releasing another update in the next day or two, v323, which raises the skin tone luma to be more accurate. The colours are accurate with v308, but skin luma is a bit low compared to where it should be, hence the update.

Hopefully then the 'sunburnt' problem will be no more!

Cheers,

Paul :)
Click to expand...

I never saw a blue tint. For me, it pulled out the unwanted orange tint from skin tones and blacks and put the balance right where it should be.
 
Batutta said:
I never saw a blue tint. For me, it pulled out the unwanted orange tint from skin tones and blacks and put the balance right where it should be.
Click to expand...

My primary target has to be accuracy over everything else :) I think I'm really there with v323.

What I will look at once all the LUTs are out though, is a Quickie Pack designed around skin tones, so that people can have the choice to get back to a place like v201 while still having accurate colour as the first stop.

Here's v323 for comparison to the previous versions, again with just the Basic Balanced Quickie v2 added to give it a nice pre-curve. It's a good image to show skin, as it has the Amazonian woman front and center, but also a fair-skinned woman in the background, so you can compare.

v323.BB3.Sample1.jpg


Cheers,

Paul :)
 
Last edited:
Been trying out v323, and I'm not seeing enough of a difference between the LUT and no LUT at all. Still vastly prefer v201. Skin looks more accurate to me and it's a better starting place for a grade. I realize you have other goals for this LUT, but for me, the only thing that kept me from using CineD was the salmon/orange skin tones, which v201 completely fixed. I'll be using that, unless another update looks better.
 
Batutta said:
Been trying out v323, and I'm not seeing enough of a difference between the LUT and no LUT at all. Still vastly prefer v201. Skin looks more accurate to me and it's a better starting place for a grade. I realize you have other goals for this LUT, but for me, the only thing that kept me from using CineD was the salmon/orange skin tones, which v201 completely fixed. I'll be using that, unless another update looks better.
Click to expand...
Of course you can use any of the LUTs available if you want :)

As far as chart accuracy goes, v323 is the most accurate for sure.

Maybe once I'm done with corrective LUTs for the various cameras still to come, I'll make a Quickie Pack with some variation based on things like you mention, to give people options :) Do try the current v323 with the Skin Pop or Skin Rostok Quickies tough - you might like the results.

Cheers,

Paul :)
 
Still from yesterday's music video shoot.

Shot with the Panasonic GH4, ETTR Cine-D as per my guide, 8 bit 4:2:0 internal, Metabones Speed Booster, Carl Zeiss 85mm/1.4 @ f4.5, 1/50th shutter, ISO320, 4K.

Skin tones via Leeming LUT One v323 plus Basic Contrasty v2 Quickie.

Any questions? :D

Lisa_Arata_Grab.jpg
 
Paul,
This is definitely the most accurate version to date. This is a really great starting point to then work with other grading tools (including finishing LUT's.)

I vote strongly in favor of keeping the approach of a separate input LUT (correction) and any number of output LUT's (finishing to taste.)

leeming_skingrade.png
 
Amazing images. I have been experiencing great success as well with v201 and will be trying v323 very soon. Outstanding work Paul. Thank you very much for all your hard work.
 
joe12south said:
Paul,
This is definitely the most accurate version to date. This is a really great starting point to then work with other grading tools (including finishing LUT's.)

I vote strongly in favor of keeping the approach of a separate input LUT (correction) and any number of output LUT's (finishing to taste.)
Click to expand...
Stunning image Joe!

And I absolutely agree with you :) Leeming LUT One will always be a corrective LUT to get to CORRECT colours first. The Quickies are the second half, where the creative tastes etc will come into play; they will always be complimentary, but never necessary.

Cheers,

Paul :)
 
I've never been happy with how GH4 footage responded to LUT's designed for "stock" REC709 footage. Using the Cine-D+Leeming LUT appears to create a much better "base" REC709 image that is more compatible with the countless stylized output LUT's floating around out there.

Leeming v323
leeming.png


Leeming + Osiris LUT
leeming_Osiris.png
 
PS. Paul, I would argue that the green/magenta axis is still undersaturated, but the balance of saturation around the wheel is definitely better than in the past. Hue is quite good across the board...it's REALLY hard for me resist the urge to try to pin everything exactly on the scope, but the results are so good right now that I'm able to tell myself, "Eh, don't sweat it. It looks great."
 
joe12south said:
I've never been happy with how GH4 footage responded to LUT's designed for "stock" REC709 footage. Using the Cine-D+Leeming LUT appears to create a much better "base" REC709 image that is more compatible with the countless stylized output LUT's floating around out there.
Click to expand...
Wow, Osiris really does a number on the curve. Almost inverted by the looks of it. I much prefer the image with just my LUT applied.

Regarding saturations on the green/magenta line, I actually felt that both were still a bit oversaturated(!), but I left them as my LUT software was telling me they were spot on.

But yes, v323 is basically as accurate as it's going to get. There's really nowhere else to go!

Your color chart image is adorable by the way! :D "All I want for Christmas is a baby and a color chart!"

Cheers,

Paul :)
 
Wow, Osiris really does a number on the curve. Almost inverted by the looks of it. I much prefer the image with just my LUT applied.
Click to expand...

Yeah, it's a very stylized look...I just wanted to demonstrate how the raw image + Leeming input LUT is "compatible" with a popular off-the-shelf LUT.
 
Paul,
Overall I think it was a wise decision to move further away from a "flat" 8-bit starting point, but after some more testing I find myself questioning if SAT 0 is a bit too hot. I say this because I find it is not hard to clip one or more channels before others at this setting. If saturation is nudged down a bit in camera ( -2 for example ) this seems much less likely, while still yielding a very natural, full-bodied image.
 
joe12south said:
Paul,
Overall I think it was a wise decision to move further away from a "flat" 8-bit starting point, but after some more testing I find myself questioning if SAT 0 is a bit too hot. I say this because I find it is not hard to clip one or more channels before others at this setting. If saturation is nudged down a bit in camera ( -2 for example ) this seems much less likely, while still yielding a very natural, full-bodied image.
Click to expand...
I can do some testing later to see if it's worth the shift. The downside of less initial saturation is that the sensor noise as a percentage/ratio of overall colour will increase, leading to perceptively noisier pictures.

I would also have to test whether it's combined with clipping at the highlight end only, or whether it is possible to clip a super saturated colour lower down the luma scale.

But if results showed that -1 or -2 were to present less chance of colour clipping without an associated increase in relative sensor noise, it might be worth recommending the change as a baseline.

Cheers!

Paul :)
 
Paul, the channel clipping I'm noticing is not in the highlights, but rather snack in the middle (chart "memory" colors). I don't think a tick or two down is going to have a meaningful impact on the noise floor, but you would have to re-evaluate the balance. I suspect this why I see green/magenta under served because orange/blue is a bit off the charts.
 
visceralpsyche said:
Wow, Osiris really does a number on the curve. Almost inverted by the looks of it. I much prefer the image with just my LUT applied.

Regarding saturations on the green/magenta line, I actually felt that both were still a bit oversaturated(!), but I left them as my LUT software was telling me they were spot on.

But yes, v323 is basically as accurate as it's going to get. There's really nowhere else to go!

Your color chart image is adorable by the way! :D "All I want for Christmas is a baby and a color chart!"

Cheers,

Paul :)
Click to expand...

Is v323 really the most accurate it can be if the color channels for a white swatch show three different values? I am using a Datacolor brand chart but even white objects in the shot show a white that has three different values in the color channels. Granted I am white balancing with this Datacolor chart which could be throwing the white color channels off slightly compared to the X-Rite brand chart. Are the differences between 18% gray on Datacolor that far off of 18% gray on X-rite?
 
Thomas Smet said:
Is v323 really the most accurate it can be if the color channels for a white swatch show three different values? I am using a Datacolor brand chart but even white objects in the shot show a white that has three different values in the color channels. Granted I am white balancing with this Datacolor chart which could be throwing the white color channels off slightly compared to the X-Rite brand chart. Are the differences between 18% gray on Datacolor that far off of 18% gray on X-rite?
Click to expand...
Well, the actual reference values don't have equal RGB values anyway (see the chart below for the actual RGB values of the X-Rite Nov.2014 reference, which is different to the Datacolor reference), but at some point it's down to the inherent error variation in the whole system. Did you white balance one second before shooting? Did a cloud pass overhead, changing the light? Shadows? Movement? Off-camera colour reflections from the surfaces surrounding the chart itself, like grass? How about exact exposure from ETTR so that LUT application exactly pegs the values at their respective luminances? These are all errors that can creep into the most accurately white balanced shot.

There might be slight variations in each individual chart too. Different dates of manufacture. Etc etc.

In short, I think v323 is as close as I'm likely to get to perfect colour. Any changes would be futzing around the edges and for every person who decides it's closer, another will say it's further away from the reference. Witness the debate between those who love v201 and consider v323 a step backwards, and those who hated v201 for its blue cast and were relieved that v323 was released.

I can say this much with certainty - v323 is by all objective measures the most accurate of all the GH4/G7 LUTs I've built, insofar as colourimetry and relative saturations and luma intensities go. Is it to everyone's taste? No. But it's objective, and based on my newly purchased, fresh X-Rite reference chart, in ideal capture conditions. Taste is subjective, and that's why I have Quickies to offer people who want to apply a more finished look to the images. There will be more Quickies in the future. For now, v323 is the best the GH4 is going to get I think, as an objective, corrective LUT to start from :)

Leeming_LUT_One_X-Rite_Reference_Chart_Nov_2014_Version.png
 
You must log in or register to reply here.
Back
Top